neulus/cloud-removal
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

things

Browse Source
This commit is contained in:
neulus
2025-11-02 22:53:04 +09:00
parent e51017897d
commit ca589fab7c
13 changed files with 888 additions and 96 deletions
Download Patch File Download Diff File Expand all files Collapse all files

6
main.py
View File

@@ -18,11 +18,11 @@ train_dataset, test_dataset = get_dataset()
device = "cuda:1"
batch_size = 8 * 4 * 2
batch_size = 32
accumulation_steps = 2
total_epoch = 500
steps_per_epoch = len(train_dataset) // batch_size
steps_per_epoch = len(train_dataset) // (batch_size)
total_steps = steps_per_epoch * total_epoch
warmup_steps = int(0.05 * total_steps)
@@ -160,7 +160,7 @@ for epoch in range(start_epoch, total_epoch):
)
wandb.log({"viz/decoded": images})
psnr, ssim, lpips = benchmark(image.cpu(), original.cpu())
psnr, ssim, lpips, flawed_lpips = benchmark(image.cpu(), original.cpu())
psnr_sum += psnr.sum().item()
ssim_sum += ssim.sum().item()
lpips_sum += lpips.sum().item()

6
pyproject.toml
View File

@@ -22,7 +22,11 @@ dependencies = [
"tqdm>=4.67.1",
"transformers>=4.56.2",
"wandb[media]>=0.22.0",
"flash-attn"
"flash-attn",
"accelerate>=1.11.0",
"hdit>=0.0.1b6",
"lovely-tensors>=0.1.19",
"natten==0.21.1+torch280cu128",
]
[tool.uv.extra-build-dependencies]

2
src/benchmark/__init__.py
View File

@@ -7,6 +7,7 @@ from torchmetrics.image import (
psnr = PeakSignalNoiseRatio(1.0, reduction="none", dim=(1, 2, 3))
lp = lpips.LPIPS(net="alex")
flawed_lp = lpips.LPIPS(net="alex")
def benchmark(image1, image2):
@@ -19,4 +20,5 @@ def benchmark(image1, image2):
size_average=False,
),
lp(image1 * 2 - 1, image2 * 2 - 1),
flawed_lp(image1 * 255, image2 * 255),
)

43
src/dataset/cuhk_cr2.py
View File

@@ -2,9 +2,11 @@ import os
from pathlib import Path
from datasets import Dataset, DatasetDict, Image
from src.dataset.preprocess import make_transform
from src.dataset.preprocess import make_nir_transform, make_transform
transform = make_transform(256)
nir_transform = make_nir_transform(256)
def get_dataset() -> tuple[Dataset, Dataset]:
@@ -13,30 +15,43 @@ def get_dataset() -> tuple[Dataset, Dataset]:
return dataset["train"], dataset["test"]
data_dir = Path("/data2/C-CUHK/CUHK-CR2")
nir_dir = Path("/data2/C-CUHK/nir/CUHK-CR2")
train_cloud = sorted((data_dir / "train/cloud").glob("*.png"))
train_cloud_nir = sorted((nir_dir / "train/cloud").glob("*.png"))
train_no_cloud = sorted((data_dir / "train/label").glob("*.png"))
train_no_cloud_nir = sorted((nir_dir / "train/label").glob("*.png"))
test_cloud = sorted((data_dir / "test/cloud").glob("*.png"))
test_cloud_nir = sorted((nir_dir / "test/cloud").glob("*.png"))
test_no_cloud = sorted((data_dir / "test/label").glob("*.png"))
test_no_cloud_nir = sorted((nir_dir / "test/label").glob("*.png"))
dataset = DatasetDict(
{
"train": Dataset.from_dict(
{
"cloud": [str(p) for p in train_cloud],
"cloud_nir": [str(p) for p in train_cloud_nir],
"label": [str(p) for p in train_no_cloud],
"label_nir": [str(p) for p in train_no_cloud_nir],
}
)
.cast_column("cloud", Image())
.cast_column("label", Image()),
.cast_column("label", Image())
.cast_column("cloud_nir", Image())
.cast_column("label_nir", Image()),
"test": Dataset.from_dict(
{
"cloud": [str(p) for p in test_cloud],
"cloud_nir": [str(p) for p in test_cloud_nir],
"label": [str(p) for p in test_no_cloud],
"label_nir": [str(p) for p in test_no_cloud_nir],
}
)
.cast_column("cloud", Image())
.cast_column("label", Image()),
.cast_column("label", Image())
.cast_column("cloud_nir", Image())
.cast_column("label_nir", Image()),
}
)
dataset = dataset.map(
@@ -45,7 +60,7 @@ def get_dataset() -> tuple[Dataset, Dataset]:
batch_size=32,
remove_columns=dataset["train"].column_names,
)
dataset.set_format(type="torch", columns=["cloud", "gt"])
dataset.set_format(type="torch", columns=["cloud", "gt", "cloud_nir", "gt_nir"])
dataset.save_to_disk("datasets/CUHK-CR2")
return dataset["train"], dataset["test"]
@@ -54,9 +69,25 @@ def get_dataset() -> tuple[Dataset, Dataset]:
def preprocess_function(examples):
x0_list = []
x1_list = []
for x0_img, x1_img in zip(examples["cloud"], examples["label"]):
x0_nir_list = []
x1_nir_list = []
for x0_img, x1_img, x0_nir, x1_nir in zip(
examples["cloud"],
examples["label"],
examples["cloud_nir"],
examples["label_nir"],
):
x0_transformed = transform(x0_img)
x1_transformed = transform(x1_img)
x0_nir = nir_transform(x0_nir)
x1_nir = nir_transform(x1_nir)
x0_list.append(x0_transformed)
x1_list.append(x1_transformed)
return {"cloud": x0_list, "gt": x1_list}
x0_nir_list.append(x0_nir)
x1_nir_list.append(x1_nir)
return {
"cloud": x0_list,
"gt": x1_list,
"cloud_nir": x0_nir_list,
"gt_nir": x1_nir_list,
}

11
src/dataset/preprocess.py
View File

@@ -14,13 +14,20 @@ def make_transform(resize_size: int = 256):
return v2.Compose([to_tensor, resize, to_float, normalize])
def make_nir_transform(resize_size: int = 256):
to_tensor = v2.ToImage()
resize = v2.Resize((resize_size, resize_size), antialias=True)
to_float = v2.ToDtype(torch.float32, scale=True)
return v2.Compose([to_tensor, v2.Grayscale(), resize, to_float])
def denormalize(tensor: torch.Tensor) -> torch.Tensor:
mean = torch.tensor([0.430, 0.411, 0.296]).view(3, 1, 1).to(tensor.device)
std = torch.tensor([0.213, 0.156, 0.143]).view(3, 1, 1).to(tensor.device)
return tensor * std + mean
return tensor[:, :3] * std + mean
def normalize(tensor: torch.Tensor) -> torch.Tensor:
mean = torch.tensor([0.430, 0.411, 0.296]).view(3, 1, 1).to(tensor.device)
std = torch.tensor([0.213, 0.156, 0.143]).view(3, 1, 1).to(tensor.device)
return (tensor - mean) / std
return (tensor[:, :3] - mean) / std

5
src/model/dino.py
View File

@@ -165,7 +165,10 @@ class DINOv3ViTRopePositionEmbedding(nn.Module):
)
angles = (
2 * math.pi * patch_coords[:, :, None] * self.inv_freq[None, None, :] # type: ignore
2
* math.pi
* patch_coords[:, :, None]
* self.inv_freq[None, None, :].to(patch_coords.device) # type: ignore
)
angles = angles.flatten(1, 2)
angles = angles.tile(2)

0
src/model/hdit.py Normal file
View File

163
src/model/utransformer.py
View File

@@ -1,4 +1,5 @@
import math
from functools import lru_cache
from typing import Optional
import torch
@@ -260,22 +261,113 @@ class DinoV3ViTDecoder(nn.Module):
# return x
class NerfEmbedder(nn.Module):
def __init__(self, in_channels, hidden_size_input, max_freqs):
super().__init__()
self.max_freqs = max_freqs
self.hidden_size_input = hidden_size_input
self.embedder = nn.Sequential(
nn.Linear(in_channels + max_freqs**2, hidden_size_input, bias=True),
)
@lru_cache
def fetch_pos(self, patch_size, device, dtype):
pos_x = torch.linspace(0, 1, patch_size, device=device, dtype=dtype)
pos_y = torch.linspace(0, 1, patch_size, device=device, dtype=dtype)
pos_y, pos_x = torch.meshgrid(pos_y, pos_x, indexing="ij")
pos_x = pos_x.reshape(-1, 1, 1)
pos_y = pos_y.reshape(-1, 1, 1)
freqs = torch.linspace(
0, self.max_freqs, self.max_freqs, dtype=dtype, device=device
)
freqs_x = freqs[None, :, None]
freqs_y = freqs[None, None, :]
coeffs = (1 + freqs_x * freqs_y) ** -1
dct_x = torch.cos(pos_x * freqs_x * torch.pi)
dct_y = torch.cos(pos_y * freqs_y * torch.pi)
dct = (dct_x * dct_y * coeffs).view(1, -1, self.max_freqs**2)
return dct
def forward(self, inputs):
target_dtype = self.embedder[0].weight.dtype
inputs = inputs.to(dtype=target_dtype)
B, P2, C = inputs.shape
patch_size = int(P2**0.5)
device = inputs.device
dtype = inputs.dtype
dct = self.fetch_pos(patch_size, device, dtype)
dct = dct.repeat(B, 1, 1)
inputs = torch.cat([inputs, dct], dim=-1)
inputs = self.embedder(inputs)
return inputs
class NerfBlock(nn.Module):
def __init__(self, hidden_size_s: int, hidden_size_x: int, mlp_ratio: int = 4):
super().__init__()
self.param_generator1 = nn.Sequential(
nn.Linear(hidden_size_s, 2 * hidden_size_x**2 * mlp_ratio, bias=True),
)
self.norm = nn.RMSNorm(hidden_size_x, eps=1e-6)
self.mlp_ratio = mlp_ratio
def forward(self, x, s):
batch_size, num_x, hidden_size_x = x.shape
mlp_params1 = self.param_generator1(s)
fc1_param1, fc2_param1 = mlp_params1.chunk(2, dim=-1)
fc1_param1 = fc1_param1.view(
batch_size, hidden_size_x, hidden_size_x * self.mlp_ratio
)
fc2_param1 = fc2_param1.view(
batch_size, hidden_size_x * self.mlp_ratio, hidden_size_x
)
# normalize fc1
normalized_fc1_param1 = torch.nn.functional.normalize(fc1_param1, dim=-2)
# normalize fc2
normalized_fc2_param1 = torch.nn.functional.normalize(fc2_param1, dim=-2)
# mlp 1
res_x = x
x = self.norm(x)
x = torch.bmm(x, normalized_fc1_param1)
x = torch.nn.functional.silu(x)
x = torch.bmm(x, normalized_fc2_param1)
x = x + res_x
return x
class NerfFinalLayer(nn.Module):
def __init__(self, hidden_size, out_channels):
super().__init__()
self.norm = nn.RMSNorm(hidden_size, eps=1e-6)
self.linear = nn.Linear(hidden_size, out_channels, bias=True)
def forward(self, x):
x = self.norm(x)
x = self.linear(x)
return x
class UTransformer(nn.Module):
def __init__(
self, config: DINOv3ViTConfig, num_classes: int, scale_factor: int = 4
self,
config: DINOv3ViTConfig,
num_classes: int,
nerf_patch=16,
nerf_hidden=64,
scale_factor: int = 4,
):
super().__init__()
self.config = config
self.scale_factor = scale_factor
self.nerf_patch_size = nerf_patch
assert config.num_hidden_layers % scale_factor == 0
self.embeddings = DINOv3ViTEmbeddings(config)
self.rope_embeddings = DINOv3ViTRopePositionEmbedding(config)
self.t_embedder = TimestepEmbedder(config.hidden_size)
# self.y_embedder = LabelEmbedder(
# num_classes, config.hidden_size, config.drop_path_rate
# ) # disable cond for now
self.encoder_layers = nn.ModuleList(
[
@@ -302,8 +394,13 @@ class UTransformer(nn.Module):
self.rest_decoder = nn.ModuleList(
[DinoConditionedLayer(config, False) for _ in range(4)]
)
self.decoder_norm = nn.LayerNorm(config.hidden_size, eps=config.layer_norm_eps)
self.decoder = DinoV3ViTDecoder(config)
# nerf!
self.nerf_encoder = NerfEmbedder(3, nerf_hidden, 8) # (rgb, hidden, freq)
self.nerf_decoder = nn.ModuleList(
[NerfBlock(self.config.hidden_size, nerf_hidden) for _ in range(12)]
)
self.final_layer = NerfFinalLayer(nerf_hidden, 3)
# freeze pretrained
self.embeddings.requires_grad_(False)
@@ -321,6 +418,13 @@ class UTransformer(nn.Module):
if time.dim() == 0:
time = time.repeat(pixel_values.shape[0])
# resolution config
B = pixel_values.shape[0]
dino_h = pixel_values.shape[-2] // self.config.patch_size
dino_w = pixel_values.shape[-1] // self.config.patch_size
nerf_h = pixel_values.shape[-2] // self.nerf_patch_size
nerf_w = pixel_values.shape[-1] // self.nerf_patch_size
pixel_values = pixel_values.to(self.embeddings.patch_embeddings.weight.dtype)
position_embeddings = self.rope_embeddings(pixel_values)
@@ -367,11 +471,52 @@ class UTransformer(nn.Module):
attention_mask=layer_head_mask,
position_embeddings=position_embeddings,
do_condition=False,
) # (batch, image // patch^2, 1024)
x = x[:, 1 + self.config.num_register_tokens :, :]
nerf_cond = nn.functional.silu(t + x) # (batch, image // patch^2, 1024)
nerf_cond = nerf_cond.reshape(
B, dino_h, dino_w, self.config.hidden_size
).permute(0, 3, 1, 2) # (batch, 1024, image // patch, image // patch)
# nerf_cond = nn.functional.interpolate(
# nerf_cond, size=(nerf_h, nerf_w), mode="bilinear", align_corners=False
# )
nerf_cond = (
nerf_cond.permute(0, 2, 3, 1)
.reshape(-1, nerf_h * nerf_w, self.config.hidden_size)
.view(-1, self.config.hidden_size)
)
x = self.decoder_norm(x)
# nerf
x_nerf = nn.functional.unfold(
pixel_values, self.nerf_patch_size, stride=self.nerf_patch_size
).transpose(1, 2)
x_nerf = x_nerf.reshape(
B * x_nerf.shape[1], -1, self.nerf_patch_size**2
).transpose(1, 2)
x_nerf = self.nerf_encoder(x_nerf)
return self.decoder(x, image_size=pixel_values.shape[-2:]), residual
for module in self.nerf_decoder:
x_nerf = module(x_nerf, nerf_cond)
x_nerf = self.final_layer(x_nerf)
num_patches = nerf_h * nerf_w
x_nerf = x_nerf.reshape(
B * num_patches, -1
) # (B*num_patches, 48): flatten pixels+RGB per patch
x_nerf = (
x_nerf.view(B, num_patches, -1).transpose(1, 2).contiguous()
) # (B, 48, num_patches)
res = nn.functional.fold(
x_nerf,
(pixel_values.shape[-2], pixel_values.shape[-1]),
kernel_size=self.nerf_patch_size,
stride=self.nerf_patch_size,
)
return res
def get_residual(
self,
@@ -410,7 +555,7 @@ class UTransformer(nn.Module):
@staticmethod
def from_pretrained_backbone(name: str):
config = DINOv3ViTConfig.from_pretrained(name)
instance = UTransformer(config, 0).to("cuda:1")
instance = UTransformer(config, 0)
weight_dict = {}
with safe_open(

19
src/rf.py
View File

@@ -22,7 +22,7 @@ use_lecam = True
class RF:
def __init__(self, model, fm="otcfm", loss="mse"):
def __init__(self, model, fm="otcfm", loss="mse", lp=None):
self.model = model
self.loss = loss
self.iter = 0
@@ -40,7 +40,7 @@ class RF:
raise NotImplementedError(
f"Unknown model {fm}, must be one of ['otcfm', 'icfm', 'fm', 'si']"
)
if not lp:
self.lpips = lpips.LPIPS(net="vgg").to("cuda:1")
self.lpips2 = lpips.LPIPS(net="alex").to("cuda:1")
@@ -53,6 +53,8 @@ class RF:
weight_decay=1e-3,
betas=(0.9, 0.95),
)
else:
self.lpips = lp
def gan_loss(self, real, fake):
global lecam_beta, lecam_anchor_real_logits, lecam_anchor_fake_logits, use_lecam
@@ -105,7 +107,7 @@ class RF:
if condition:
vt = self.model(xt, t, cloud)
else:
vt, _ = self.model(xt, t)
vt = self.model(xt, t)
if self.loss == "mse":
loss = ((vt - ut) ** 2).mean(dim=list(range(1, len(gt.shape))))
@@ -116,11 +118,18 @@ class RF:
denormalize(gt) * 2 - 1,
denormalize(xt + (1 - t[:, None, None, None]) * vt) * 2 - 1,
)
ssim = 1 - ms_ssim(
denormalize(gt),
denormalize(xt + (1 - t[:, None, None, None]) * vt),
data_range=1.0,
size_average=False,
)
loss_list = {
"train/mse": mse.mean().item(),
"train/lpips": lpips.mean().item(),
"train/ssim": ssim.mean().item(),
}
loss = mse + lpips * 2.0
loss = mse + lpips * 2.0 + ssim
elif self.loss == "gan_lpips_mse":
self.iter += 1
# if self.iter % 4 == 0:
@@ -179,7 +188,7 @@ class RF:
)
else:
traj = odeint(
lambda t, x: self.model(x, t)[0],
lambda t, x: self.model(x, t),
cloud,
t_span,
rtol=tol,

3
test_dataset.py Normal file
View File

@@ -0,0 +1,3 @@
from src.dataset.cuhk_cr2 import get_dataset
get_dataset()

247
train_accelerator.py Normal file
View File

@@ -0,0 +1,247 @@
import math
import os
from datetime import timedelta
import lpips
import torch
import torch.optim as optim
from accelerate import Accelerator, InitProcessGroupKwargs
from torch.utils.data import DataLoader
from torchvision.utils import make_grid
from tqdm import tqdm
import wandb
from src.benchmark import benchmark
from src.dataset.cuhk_cr2 import get_dataset
from src.dataset.preprocess import denormalize
from src.model.utransformer import UTransformer
from src.rf import RF
# --- Configuration ---
batch_size = 16
accumulation_steps = 2
total_epoch = 500
grad_norm = 1.0
learning_rate = 3e-4
# --- Accelerator Setup ---
# Set a longer timeout for initialization, which can be useful when downloading
# large models or datasets on multiple nodes.
kwargs = InitProcessGroupKwargs(timeout=timedelta(seconds=1800))
accelerator = Accelerator(
gradient_accumulation_steps=accumulation_steps,
mixed_precision="bf16", # Use "bf16" for modern GPUs, or "fp16"
log_with="wandb",
kwargs_handlers=[kwargs],
)
# --- Dataset Loading ---
# Load datasets on the main process. They will be accessible by all processes.
train_dataset, test_dataset = get_dataset()
train_dataset, test_dataset = (
DataLoader(train_dataset, batch_size=batch_size),
DataLoader(test_dataset, batch_size=batch_size),
)
# --- LR Scheduler Logic ---
# Correctly calculate total steps based on optimizer updates, not micro-batches.
# Use math.ceil to account for the last partial batch.
num_batches_per_epoch = math.ceil(len(train_dataset) / batch_size)
optimizer_steps_per_epoch = math.ceil(num_batches_per_epoch / accumulation_steps)
total_steps = optimizer_steps_per_epoch * total_epoch
warmup_steps = int(0.05 * total_steps)
# --- Model, Optimizer, and RF Helper Initialization ---
# Initialize on CPU. Accelerator will move them to the correct device.
model = UTransformer.from_pretrained_backbone(
"facebook/dinov3-vitl16-pretrain-sat493m"
).bfloat16()
lp = lpips.LPIPS(net="vgg")
rf = RF(model, "icfm", "lpips_mse", lp) # RF holds a reference to the model
optimizer = optim.AdamW(model.parameters(), lr=learning_rate)
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=total_epoch)
# --- Prepare objects with Accelerator ---
# We prepare everything except the train_dataloader, which is created per-epoch
# to allow for webdataset's shuffling mechanism.
model, lp, optimizer, scheduler, train_dataset, test_dataset = accelerator.prepare(
model, lp, optimizer, scheduler, train_dataset, test_dataset
)
# --- W&B and Checkpoint Setup ---
# Initialize tracker (wandb) on the main process
accelerator.init_trackers(
project_name="cloud-removal-kmu",
config={
"batch_size": batch_size,
"accumulation_steps": accumulation_steps,
"total_epoch": total_epoch,
"learning_rate": learning_rate,
"grad_norm": grad_norm,
"total_steps": total_steps,
},
)
# Use the run name from the tracker for a consistent artifact path
# This check is needed in case there are no trackers configured.
run_name = "nerf-3"
if accelerator.trackers:
run_name = accelerator.trackers[0].run.name
artifact_dir = f"artifact/{run_name}"
checkpoint_dir = os.path.join(artifact_dir, "checkpoints")
if accelerator.is_main_process:
os.makedirs(checkpoint_dir, exist_ok=True)
accelerator.wait_for_everyone() # Ensure directory is created before any process tries to access it
# Register scheduler for checkpointing
accelerator.register_for_checkpointing(scheduler)
start_epoch = 0
# Check if a checkpoint exists to resume training
if os.path.exists(checkpoint_dir):
try:
accelerator.print(f"Resuming from checkpoint: {checkpoint_dir}")
accelerator.load_state(checkpoint_dir)
# Manually load the epoch from a tracker file
if os.path.exists(os.path.join(checkpoint_dir, "epoch_tracker.pt")):
start_epoch = (
torch.load(os.path.join(checkpoint_dir, "epoch_tracker.pt"))["epoch"]
+ 1
)
except Exception as e:
accelerator.print(
f"Could not load checkpoint. Starting from scratch. Error: {e}"
)
# --- Training Loop ---
for epoch in range(start_epoch, total_epoch):
model.train()
lossbin = {i: 0 for i in range(10)}
losscnt = {i: 1e-6 for i in range(10)}
progress_bar = tqdm(
train_dataset,
disable=not accelerator.is_local_main_process,
desc=f"Epoch {epoch + 1}/{total_epoch}",
)
for step, batch in enumerate(progress_bar):
cloud, gt = batch["cloud"], batch["gt"]
with accelerator.accumulate(model):
# Forward pass is automatically handled with mixed precision
loss, blsct, loss_list = rf.forward(gt, cloud)
accelerator.backward(loss)
if accelerator.sync_gradients:
accelerator.clip_grad_norm_(model.parameters(), grad_norm)
optimizer.step()
optimizer.zero_grad()
# Log metrics
if accelerator.sync_gradients:
avg_loss = accelerator.gather(loss).mean().item()
current_step = epoch * optimizer_steps_per_epoch + (
step // accumulation_steps
)
accelerator.log(
{
"train/loss": avg_loss,
"train/lr": scheduler.get_last_lr()[0],
},
)
accelerator.log(loss_list)
# This per-process logging is an approximation. For perfect accuracy,
# `blsct` would need to be gathered from all processes.
for t, lss in blsct:
bin_idx = min(int(t * 10), 9)
lossbin[bin_idx] += lss
losscnt[bin_idx] += 1
# Log epoch-level metrics from the main process
if accelerator.is_main_process:
epoch_metrics = {
f"lossbin/lossbin_{i}": lossbin[i] / losscnt[i] for i in range(10)
}
epoch_metrics["epoch"] = epoch
accelerator.log(epoch_metrics)
# --- Evaluation and Checkpointing ---
if (epoch + 1) % 50 == 0:
model.eval()
psnr_sum, ssim_sum, lpips_sum, count = 0.0, 0.0, 0.0, 0
with torch.no_grad():
for i, batch in tqdm(
enumerate(test_dataset),
disable=not accelerator.is_local_main_process,
desc=f"Benchmark {epoch + 1}/{total_epoch}",
):
images = rf.sample(batch["cloud"])
image = denormalize(images[-1]).clamp(0, 1)
original = denormalize(batch["gt"]).clamp(0, 1)
# Gather results from all processes for accurate metrics
image_gathered = accelerator.gather_for_metrics(image)
original_gathered = accelerator.gather_for_metrics(original)
if accelerator.is_main_process:
# Log visualization images from the first batch on the main process
if i == 0:
demo_images = [images[0][:4], images[-1][:4]]
for step_idx, demo in enumerate(demo_images):
grid = make_grid(
denormalize(demo).clamp(0, 1).float().cpu(), nrow=2
)
wandb_image = wandb.Image(grid, caption=f"step {step_idx}")
accelerator.log({"viz/decoded": wandb_image})
psnr, ssim, lpips, flawed_lpips = benchmark(
image_gathered.cpu(), original_gathered.cpu()
)
psnr_sum += psnr.sum().item()
ssim_sum += ssim.sum().item()
lpips_sum += lpips.sum().item()
count += image_gathered.shape[0]
accelerator.wait_for_everyone()
if accelerator.is_main_process:
avg_psnr = psnr_sum / count if count > 0 else 0
avg_ssim = ssim_sum / count if count > 0 else 0
avg_lpips = lpips_sum / count if count > 0 else 0
accelerator.log(
{
"eval/psnr": avg_psnr,
"eval/ssim": avg_ssim,
"eval/lpips": avg_lpips,
"epoch": epoch + 1,
}
)
# Save checkpoint on the main process
accelerator.save_state(os.path.join(checkpoint_dir, f"epoch_{epoch + 1}"))
accelerator.save_state(checkpoint_dir) # Overwrite latest
torch.save(
{"epoch": epoch}, os.path.join(checkpoint_dir, "epoch_tracker.pt")
)
# scheduler.step()
# --- Final Save and Cleanup ---
accelerator.wait_for_everyone()
if accelerator.is_main_process:
accelerator.print("Saving final model state.")
accelerator.save_state(checkpoint_dir)
torch.save(
{"epoch": total_epoch - 1}, os.path.join(checkpoint_dir, "epoch_tracker.pt")
)
accelerator.end_training()

265
train_accelerator_hdit.py Normal file
View File

@@ -0,0 +1,265 @@
import math
import os
from datetime import timedelta
import lpips
import torch
import torch.optim as optim
from accelerate import Accelerator, InitProcessGroupKwargs
from hdit import HDiT
from torch.utils.data import DataLoader
from torchvision.utils import make_grid
from tqdm import tqdm
import wandb
from src.benchmark import benchmark
from src.dataset.cuhk_cr2 import get_dataset
from src.dataset.preprocess import denormalize
from src.rf import RF
# --- Configuration ---
batch_size = 4
accumulation_steps = 2
total_epoch = 500
grad_norm = 1.0
learning_rate = 3e-4
# --- Accelerator Setup ---
# Set a longer timeout for initialization, which can be useful when downloading
# large models or datasets on multiple nodes.
kwargs = InitProcessGroupKwargs(timeout=timedelta(seconds=1800))
accelerator = Accelerator(
gradient_accumulation_steps=accumulation_steps,
mixed_precision="bf16", # Use "bf16" for modern GPUs, or "fp16"
log_with="wandb",
kwargs_handlers=[kwargs],
)
# --- Dataset Loading ---
# Load datasets on the main process. They will be accessible by all processes.
train_dataset, test_dataset = get_dataset()
train_dataset, test_dataset = (
DataLoader(train_dataset, batch_size=batch_size), # type: ignore
DataLoader(test_dataset, batch_size=batch_size), # type: ignore
)
num_batches_per_epoch = math.ceil(len(train_dataset) / batch_size)
optimizer_steps_per_epoch = math.ceil(num_batches_per_epoch / accumulation_steps)
total_steps = optimizer_steps_per_epoch * total_epoch
warmup_steps = int(0.05 * total_steps)
# --- Model, Optimizer, and RF Helper Initialization ---
# Initialize on CPU. Accelerator will move them to the correct device.
model = HDiT(
in_channels=4,
out_channels=4,
patch_size=[1, 1], # type: ignore
widths=[256, 512],
middle_width=1024,
depths=[4, 4],
middle_depth=8,
mapping_width=512,
mapping_depth=2,
).bfloat16()
print(sum(p.numel() for p in model.parameters() if p.requires_grad), "params")
lp = lpips.LPIPS(net="vgg")
rf = RF(model, "icfm", "lpips_mse", lp) # RF holds a reference to the model
optimizer = optim.AdamW(model.parameters(), lr=learning_rate)
scheduler = optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=total_epoch)
# --- Prepare objects with Accelerator ---
# We prepare everything except the train_dataloader, which is created per-epoch
# to allow for webdataset's shuffling mechanism.
model, lp, optimizer, scheduler, train_dataset, test_dataset = accelerator.prepare(
model, lp, optimizer, scheduler, train_dataset, test_dataset
)
# --- W&B and Checkpoint Setup ---
# Initialize tracker (wandb) on the main process
accelerator.init_trackers(
project_name="cloud-removal-kmu",
config={
"batch_size": batch_size,
"accumulation_steps": accumulation_steps,
"total_epoch": total_epoch,
"learning_rate": learning_rate,
"grad_norm": grad_norm,
"total_steps": total_steps,
},
)
# Use the run name from the tracker for a consistent artifact path
# This check is needed in case there are no trackers configured.
run_name = "testest"
if accelerator.trackers:
run_name = accelerator.trackers[0].run.name
artifact_dir = f"artifact/{run_name}"
checkpoint_dir = os.path.join(artifact_dir, "checkpoints")
if accelerator.is_main_process:
os.makedirs(checkpoint_dir, exist_ok=True)
accelerator.wait_for_everyone() # Ensure directory is created before any process tries to access it
# Register scheduler for checkpointing
accelerator.register_for_checkpointing(scheduler)
start_epoch = 0
# Check if a checkpoint exists to resume training
if os.path.exists(checkpoint_dir):
try:
accelerator.print(f"Resuming from checkpoint: {checkpoint_dir}")
accelerator.load_state(checkpoint_dir)
# Manually load the epoch from a tracker file
if os.path.exists(os.path.join(checkpoint_dir, "epoch_tracker.pt")):
start_epoch = (
torch.load(os.path.join(checkpoint_dir, "epoch_tracker.pt"))["epoch"]
+ 1
)
except Exception as e:
accelerator.print(
f"Could not load checkpoint. Starting from scratch. Error: {e}"
)
# --- Training Loop ---
for epoch in range(start_epoch, total_epoch):
model.train()
lossbin = {i: 0 for i in range(10)}
losscnt = {i: 1e-6 for i in range(10)}
progress_bar = tqdm(
train_dataset,
disable=not accelerator.is_local_main_process,
desc=f"Epoch {epoch + 1}/{total_epoch}",
)
for step, batch in enumerate(progress_bar):
cloud, gt = batch["cloud"], batch["gt"]
with accelerator.accumulate(model):
# Forward pass is automatically handled with mixed precision
loss, blsct, loss_list = rf.forward(
torch.cat((batch["gt"], batch["gt_nir"]), dim=1),
torch.cat((batch["cloud"], batch["cloud_nir"]), dim=1),
)
accelerator.backward(loss)
if accelerator.sync_gradients:
accelerator.clip_grad_norm_(model.parameters(), grad_norm)
optimizer.step()
optimizer.zero_grad()
# Log metrics
if accelerator.sync_gradients:
avg_loss = accelerator.gather(loss).mean().item() # type: ignore
current_step = epoch * optimizer_steps_per_epoch + (
step // accumulation_steps
)
accelerator.log(
{
"train/loss": avg_loss,
"train/lr": scheduler.get_last_lr()[0],
},
# step=current_step,
)
accelerator.log(loss_list)
# This per-process logging is an approximation. For perfect accuracy,
# `blsct` would need to be gathered from all processes.
for t, lss in blsct:
bin_idx = min(int(t * 10), 9)
lossbin[bin_idx] += lss
losscnt[bin_idx] += 1
# Log epoch-level metrics from the main process
if accelerator.is_main_process:
epoch_metrics = {
f"lossbin/lossbin_{i}": lossbin[i] / losscnt[i] for i in range(10)
}
epoch_metrics["epoch"] = epoch
accelerator.log(epoch_metrics)
# --- Evaluation and Checkpointing ---
if (epoch + 1) % 50 == 0:
model.eval()
psnr_sum, ssim_sum, lpips_sum, flawed_lpips_sum, count = 0.0, 0.0, 0.0, 0.0, 0
with torch.no_grad():
for i, batch in tqdm(
enumerate(test_dataset),
disable=not accelerator.is_local_main_process,
desc=f"Benchmark {epoch + 1}/{total_epoch}",
):
images = rf.sample(
torch.cat((batch["cloud"], batch["cloud_nir"]), dim=1)
)
image = denormalize(images[-1]).clamp(0, 1)
original = denormalize(batch["gt"]).clamp(0, 1)
# Gather results from all processes for accurate metrics
image_gathered = accelerator.gather_for_metrics(image)
original_gathered = accelerator.gather_for_metrics(original)
if accelerator.is_main_process:
# Log visualization images from the first batch on the main process
if i == 0:
demo_images = [images[0][:4], images[-1][:4]]
for step_idx, demo in enumerate(demo_images):
grid = make_grid(
denormalize(demo).clamp(0, 1).float().cpu(), nrow=2
)
wandb_image = wandb.Image(grid, caption=f"step {step_idx}")
accelerator.log({"viz/decoded": wandb_image})
psnr, ssim, lpips, flawed_lpips = benchmark(
image_gathered.cpu(), # type: ignore
original_gathered.cpu(), # type: ignore
)
psnr_sum += psnr.sum().item()
ssim_sum += ssim.sum().item()
lpips_sum += lpips.sum().item()
flawed_lpips_sum += flawed_lpips.sum().item()
count += image_gathered.shape[0] # type: ignore
accelerator.wait_for_everyone()
if accelerator.is_main_process:
avg_psnr = psnr_sum / count if count > 0 else 0
avg_ssim = ssim_sum / count if count > 0 else 0
avg_lpips = lpips_sum / count if count > 0 else 0
avg_flawed_lpips = flawed_lpips_sum / count if count > 0 else 0
accelerator.log(
{
"eval/psnr": avg_psnr,
"eval/ssim": avg_ssim,
"eval/lpips": avg_lpips,
"eval/flawed_lpips": avg_flawed_lpips,
"epoch": epoch + 1,
}
)
# Save checkpoint on the main process
accelerator.save_state(os.path.join(checkpoint_dir, f"epoch_{epoch + 1}"))
accelerator.save_state(checkpoint_dir) # Overwrite latest
torch.save(
{"epoch": epoch}, os.path.join(checkpoint_dir, "epoch_tracker.pt")
)
# scheduler.step()
# --- Final Save and Cleanup ---
accelerator.wait_for_everyone()
if accelerator.is_main_process:
accelerator.print("Saving final model state.")
accelerator.save_state(checkpoint_dir)
torch.save(
{"epoch": total_epoch - 1}, os.path.join(checkpoint_dir, "epoch_tracker.pt")
)
accelerator.end_training()

190
uv.lock generated
View File

@@ -15,6 +15,24 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/8f/aa/ba0014cc4659328dc818a28827be78e6d97312ab0cb98105a770924dc11e/absl_py-2.3.1-py3-none-any.whl", hash = "sha256:eeecf07f0c2a93ace0772c92e596ace6d3d3996c042b2128459aaae2a76de11d", size = 135811, upload-time = "2025-07-03T09:31:42.253Z" },
]
[[package]]
name = "accelerate"
version = "1.11.0"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "huggingface-hub" },
{ name = "numpy" },
{ name = "packaging" },
{ name = "psutil" },
{ name = "pyyaml" },
{ name = "safetensors" },
{ name = "torch" },
]
sdist = { url = "https://files.pythonhosted.org/packages/23/60/2757c4f03a8705dbf80b1268b03881927878dca5ed07d74f733fb6c219e0/accelerate-1.11.0.tar.gz", hash = "sha256:bb1caf2597b4cd632b917b5000c591d10730bb024a79746f1ee205bba80bd229", size = 393715, upload-time = "2025-10-20T14:42:25.025Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/77/85/85951bc0f9843e2c10baaa1b6657227056095de08f4d1eea7d8b423a6832/accelerate-1.11.0-py3-none-any.whl", hash = "sha256:a628fa6beb069b8e549460fc449135d5bd8d73e7a11fd09f0bc9fc4ace7f06f1", size = 375777, upload-time = "2025-10-20T14:42:23.256Z" },
]
[[package]]
name = "aiohappyeyeballs"
version = "2.6.1"
@@ -295,10 +313,14 @@ name = "cloud-removal"
version = "0.1.0"
source = { virtual = "." }
dependencies = [
{ name = "accelerate" },
{ name = "datasets" },
{ name = "einops" },
{ name = "flash-attn" },
{ name = "hdit" },
{ name = "lovely-tensors" },
{ name = "lpips" },
{ name = "natten" },
{ name = "pyright" },
{ name = "python-lsp-server" },
{ name = "pytorch-msssim" },
@@ -317,10 +339,14 @@ dependencies = [
[package.metadata]
requires-dist = [
{ name = "accelerate", specifier = ">=1.11.0" },
{ name = "datasets", specifier = ">=4.1.1" },
{ name = "einops", specifier = ">=0.8.1" },
{ name = "flash-attn" },
{ name = "hdit", specifier = ">=0.0.1b6" },
{ name = "lovely-tensors", specifier = ">=0.1.19" },
{ name = "lpips", specifier = ">=0.1.4" },
{ name = "natten", specifier = "==0.21.1+torch280cu128" },
{ name = "pyright", specifier = ">=1.1.405" },
{ name = "python-lsp-server", specifier = ">=1.13.1" },
{ name = "pytorch-msssim", specifier = ">=1.0.0" },
@@ -454,6 +480,20 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/f4/c8/09012ac195a0aab58755800d2efdc0e7d5905053509f12cb5d136c911cda/datasets-4.1.1-py3-none-any.whl", hash = "sha256:62e4f6899a36be9ec74a7e759a6951253cc85b3fcfa0a759b0efa8353b149dac", size = 503623, upload-time = "2025-09-18T13:14:25.111Z" },
]
[[package]]
name = "dctorch"
version = "0.1.2"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "numpy" },
{ name = "scipy" },
{ name = "torch" },
]
sdist = { url = "https://files.pythonhosted.org/packages/76/db/75bef6df7b6de66a9291a4924f9b836a60ff53da0937c84aa44c23fbac7d/dctorch-0.1.2.tar.gz", hash = "sha256:4e67cb0248a87eb40d598061225638cda7db64f8057ed2120869683bfae5106e", size = 1767, upload-time = "2023-09-21T23:39:38.083Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/cd/47/61b627404c2d6f31dcbc491ff83da1f4336c7ae7893cfdc6c52db490ec59/dctorch-0.1.2-py3-none-any.whl", hash = "sha256:ce1dde9e3aabb941240e43522212d2e03dddae81279968c8ab182c59a5b41c8f", size = 2278, upload-time = "2023-09-21T23:39:36.584Z" },
]
[[package]]
name = "debugpy"
version = "1.8.17"
@@ -506,6 +546,18 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/56/7b/af3d0da15bed3a8665419bb3a630585756920f4ad67abfdfef26240ebcc0/docstring_to_markdown-0.17-py3-none-any.whl", hash = "sha256:fd7d5094aa83943bf5f9e1a13701866b7c452eac19765380dead666e36d3711c", size = 23479, upload-time = "2025-05-02T15:09:06.676Z" },
]
[[package]]
name = "einop"
version = "0.0.1"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "einops" },
]
sdist = { url = "https://files.pythonhosted.org/packages/04/7b/99e9939fd089dff5d5f8d3030ca04b5b30a4286efd483da0f2e07a5db90e/einop-0.0.1.tar.gz", hash = "sha256:15add3732a54eb6c49618be0c59b97163e46ae87aeb5e387d9c418402ea33178", size = 3132, upload-time = "2022-03-07T21:28:32.025Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/c5/ea/45cf629f41dcbe1da97080dd254a9bf344f39baa53c72eb417666b2100bc/einop-0.0.1-py3-none-any.whl", hash = "sha256:edff8c3916cf4a22eccd31254d752ec7959388bd6877eb5285eed6ca5170e36b", size = 3040, upload-time = "2022-03-07T21:28:29.864Z" },
]
[[package]]
name = "einops"
version = "0.8.1"
@@ -524,6 +576,18 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/c1/ea/53f2148663b321f21b5a606bd5f191517cf40b7072c0497d3c92c4a13b1e/executing-2.2.1-py2.py3-none-any.whl", hash = "sha256:760643d3452b4d777d295bb167ccc74c64a81df23fb5e08eff250c425a4b2017", size = 28317, upload-time = "2025-09-01T09:48:08.5Z" },
]
[[package]]
name = "fastcore"
version = "1.8.14"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "packaging" },
]
sdist = { url = "https://files.pythonhosted.org/packages/41/fc/4782041a7e96ae3de2b6bc7a287693d619688d938f43e6d9e70a23874d51/fastcore-1.8.14.tar.gz", hash = "sha256:869735ef493dbc7e5e8cbfc35fa3310772ce4c768d5b3a82d6a0d571148401be", size = 83648, upload-time = "2025-10-29T05:38:46.285Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/ed/c6/236247deaa155fad1b38b6deb95b8b76efd20f5107b4577eee42002cbf11/fastcore-1.8.14-py3-none-any.whl", hash = "sha256:a02a749c26243ffd54d6dd11165cf4a556c7cb08f4c7e47ff67b32c7b0183ce9", size = 86791, upload-time = "2025-10-29T05:38:44.343Z" },
]
[[package]]
name = "filelock"
version = "3.19.1"
@@ -682,6 +746,27 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/01/61/d4b89fec821f72385526e1b9d9a3a0385dda4a72b206d28049e2c7cd39b8/gitpython-3.1.45-py3-none-any.whl", hash = "sha256:8908cb2e02fb3b93b7eb0f2827125cb699869470432cc885f019b8fd0fccff77", size = 208168, upload-time = "2025-07-24T03:45:52.517Z" },
]
[[package]]
name = "hdit"
version = "0.0.1b6"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "dctorch" },
{ name = "einop" },
{ name = "pillow" },
{ name = "safetensors" },
{ name = "scipy" },
{ name = "torch" },
{ name = "torchdiffeq" },
{ name = "torchsde" },
{ name = "torchvision" },
{ name = "tqdm" },
]
sdist = { url = "https://files.pythonhosted.org/packages/cc/ca/6e00358e0af086661a1a816cb4a73f3202d87ffb64388e3c1ff2530168ec/hdit-0.0.1b6.tar.gz", hash = "sha256:35f723cf0955d5217fe16f8e2ad703c7dcd5ec6da7bd7b1a0f661849b58bd1b4", size = 33279, upload-time = "2025-09-10T13:28:10.316Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/d1/c6/98eb867319a46c310db5be37c842709669f596cc0a0fd0de3b18d9f66924/hdit-0.0.1b6-py3-none-any.whl", hash = "sha256:de8efd85286b5b50199dbb9306bd2de5098bf29e186e4290d164ab40b70be524", size = 36277, upload-time = "2025-09-10T13:28:09.121Z" },
]
[[package]]
name = "hf-xet"
version = "1.1.10"
@@ -995,6 +1080,34 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/de/73/3d757cb3fc16f0f9794dd289bcd0c4a031d9cf54d8137d6b984b2d02edf3/lightning_utilities-0.15.2-py3-none-any.whl", hash = "sha256:ad3ab1703775044bbf880dbf7ddaaac899396c96315f3aa1779cec9d618a9841", size = 29431, upload-time = "2025-08-06T13:57:38.046Z" },
]
[[package]]
name = "lovely-numpy"
version = "0.2.17"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "fastcore" },
{ name = "matplotlib" },
{ name = "numpy" },
{ name = "packaging" },
]
sdist = { url = "https://files.pythonhosted.org/packages/2c/5c/e422f913b5722bc0e697342cf16e7b39fe04ebf694e1334034ef913bbb59/lovely_numpy-0.2.17.tar.gz", hash = "sha256:c414db415701f060c0875cd1d8013a7167664f95e85d11644fdff646941abdb0", size = 24396, upload-time = "2025-10-23T08:16:36.334Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/ee/87/cd2c462355886debdd8ef2064e7de10fcb5eea2928c5c7a1d3566d36cc2d/lovely_numpy-0.2.17-py3-none-any.whl", hash = "sha256:17e1b817d10ce28ce7f4d943f87d4ae7c033250b03c24284af436c200b51f1ac", size = 24424, upload-time = "2025-10-23T08:16:35.024Z" },
]
[[package]]
name = "lovely-tensors"
version = "0.1.19"
source = { registry = "https://pypi.org/simple" }
dependencies = [
{ name = "lovely-numpy" },
{ name = "torch" },
]
sdist = { url = "https://files.pythonhosted.org/packages/f8/c2/8c21ea15038a2e98079285c52a180e53ea7f0aadcf5b1fa58c1cadfb68da/lovely_tensors-0.1.19.tar.gz", hash = "sha256:0a9cec41c6a13d7de3ca3688c10f61991071352116c5303c3e62c91febf32016", size = 22558, upload-time = "2025-10-04T15:56:33.085Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/a0/ae/93fd971b7a067ddca543302e7632509a36078b48ad0d08643bc17ac35cf3/lovely_tensors-0.1.19-py3-none-any.whl", hash = "sha256:b0fee4f5cfbb0494be6bbce0963bf8e39edc1e4ef89afe922c30edfb55d65237", size = 19394, upload-time = "2025-10-04T15:56:31.682Z" },
]
[[package]]
name = "lpips"
version = "0.1.4"
@@ -1239,6 +1352,16 @@ wheels = [
{ url = "https://files.pythonhosted.org/packages/50/3b/0e2c535c3e6970cfc5763b67f6cc31accaab35a7aa3e322fb6a12830450f/narwhals-2.6.0-py3-none-any.whl", hash = "sha256:3215ea42afb452c6c8527e79cefbe542b674aa08d7e2e99d46b2c9708870e0d4", size = 408435, upload-time = "2025-09-29T09:08:54.503Z" },
]
[[package]]
name = "natten"
version = "0.21.1+torch280cu128"
source = { registry = "https://whl.natten.org/" }
wheels = [
{ url = "https://github.com/SHI-Labs/NATTEN/releases/download/v0.21.1/natten-0.21.1%2Btorch280cu128-cp312-cp312-linux_x86_64.whl" },
{ url = "https://github.com/SHI-Labs/NATTEN/releases/download/v0.21.1/natten-0.21.1%2Btorch280cu128-cp313-cp313-linux_x86_64.whl" },
{ url = "https://github.com/SHI-Labs/NATTEN/releases/download/v0.21.1/natten-0.21.1%2Btorch280cu128-cp313-cp313t-linux_x86_64.whl" },
]
[[package]]
name = "nest-asyncio"
version = "1.6.0"
@@ -1268,65 +1391,18 @@ wheels = [
[[package]]
name = "numpy"
version = "2.3.3"
version = "1.26.4"
source = { registry = "https://pypi.org/simple" }
sdist = { url = "https://files.pythonhosted.org/packages/d0/19/95b3d357407220ed24c139018d2518fab0a61a948e68286a25f1a4d049ff/numpy-2.3.3.tar.gz", hash = "sha256:ddc7c39727ba62b80dfdbedf400d1c10ddfa8eefbd7ec8dcb118be8b56d31029", size = 20576648, upload-time = "2025-09-09T16:54:12.543Z" }
sdist = { url = "https://files.pythonhosted.org/packages/65/6e/09db70a523a96d25e115e71cc56a6f9031e7b8cd166c1ac8438307c14058/numpy-1.26.4.tar.gz", hash = "sha256:2a02aba9ed12e4ac4eb3ea9421c420301a0c6460d9830d74a9df87efa4912010", size = 15786129, upload-time = "2024-02-06T00:26:44.495Z" }
wheels = [
{ url = "https://files.pythonhosted.org/packages/51/5d/bb7fc075b762c96329147799e1bcc9176ab07ca6375ea976c475482ad5b3/numpy-2.3.3-cp312-cp312-macosx_10_13_x86_64.whl", hash = "sha256:cfdd09f9c84a1a934cde1eec2267f0a43a7cd44b2cca4ff95b7c0d14d144b0bf", size = 20957014, upload-time = "2025-09-09T15:56:29.966Z" },
{ url = "https://files.pythonhosted.org/packages/6b/0e/c6211bb92af26517acd52125a237a92afe9c3124c6a68d3b9f81b62a0568/numpy-2.3.3-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:cb32e3cf0f762aee47ad1ddc6672988f7f27045b0783c887190545baba73aa25", size = 14185220, upload-time = "2025-09-09T15:56:32.175Z" },
{ url = "https://files.pythonhosted.org/packages/22/f2/07bb754eb2ede9073f4054f7c0286b0d9d2e23982e090a80d478b26d35ca/numpy-2.3.3-cp312-cp312-macosx_14_0_arm64.whl", hash = "sha256:396b254daeb0a57b1fe0ecb5e3cff6fa79a380fa97c8f7781a6d08cd429418fe", size = 5113918, upload-time = "2025-09-09T15:56:34.175Z" },
{ url = "https://files.pythonhosted.org/packages/81/0a/afa51697e9fb74642f231ea36aca80fa17c8fb89f7a82abd5174023c3960/numpy-2.3.3-cp312-cp312-macosx_14_0_x86_64.whl", hash = "sha256:067e3d7159a5d8f8a0b46ee11148fc35ca9b21f61e3c49fbd0a027450e65a33b", size = 6647922, upload-time = "2025-09-09T15:56:36.149Z" },
{ url = "https://files.pythonhosted.org/packages/5d/f5/122d9cdb3f51c520d150fef6e87df9279e33d19a9611a87c0d2cf78a89f4/numpy-2.3.3-cp312-cp312-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:1c02d0629d25d426585fb2e45a66154081b9fa677bc92a881ff1d216bc9919a8", size = 14281991, upload-time = "2025-09-09T15:56:40.548Z" },
{ url = "https://files.pythonhosted.org/packages/51/64/7de3c91e821a2debf77c92962ea3fe6ac2bc45d0778c1cbe15d4fce2fd94/numpy-2.3.3-cp312-cp312-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:d9192da52b9745f7f0766531dcfa978b7763916f158bb63bdb8a1eca0068ab20", size = 16641643, upload-time = "2025-09-09T15:56:43.343Z" },
{ url = "https://files.pythonhosted.org/packages/30/e4/961a5fa681502cd0d68907818b69f67542695b74e3ceaa513918103b7e80/numpy-2.3.3-cp312-cp312-musllinux_1_2_aarch64.whl", hash = "sha256:cd7de500a5b66319db419dc3c345244404a164beae0d0937283b907d8152e6ea", size = 16056787, upload-time = "2025-09-09T15:56:46.141Z" },
{ url = "https://files.pythonhosted.org/packages/99/26/92c912b966e47fbbdf2ad556cb17e3a3088e2e1292b9833be1dfa5361a1a/numpy-2.3.3-cp312-cp312-musllinux_1_2_x86_64.whl", hash = "sha256:93d4962d8f82af58f0b2eb85daaf1b3ca23fe0a85d0be8f1f2b7bb46034e56d7", size = 18579598, upload-time = "2025-09-09T15:56:49.844Z" },
{ url = "https://files.pythonhosted.org/packages/17/b6/fc8f82cb3520768718834f310c37d96380d9dc61bfdaf05fe5c0b7653e01/numpy-2.3.3-cp312-cp312-win32.whl", hash = "sha256:5534ed6b92f9b7dca6c0a19d6df12d41c68b991cef051d108f6dbff3babc4ebf", size = 6320800, upload-time = "2025-09-09T15:56:52.499Z" },
{ url = "https://files.pythonhosted.org/packages/32/ee/de999f2625b80d043d6d2d628c07d0d5555a677a3cf78fdf868d409b8766/numpy-2.3.3-cp312-cp312-win_amd64.whl", hash = "sha256:497d7cad08e7092dba36e3d296fe4c97708c93daf26643a1ae4b03f6294d30eb", size = 12786615, upload-time = "2025-09-09T15:56:54.422Z" },
{ url = "https://files.pythonhosted.org/packages/49/6e/b479032f8a43559c383acb20816644f5f91c88f633d9271ee84f3b3a996c/numpy-2.3.3-cp312-cp312-win_arm64.whl", hash = "sha256:ca0309a18d4dfea6fc6262a66d06c26cfe4640c3926ceec90e57791a82b6eee5", size = 10195936, upload-time = "2025-09-09T15:56:56.541Z" },
{ url = "https://files.pythonhosted.org/packages/7d/b9/984c2b1ee61a8b803bf63582b4ac4242cf76e2dbd663efeafcb620cc0ccb/numpy-2.3.3-cp313-cp313-macosx_10_13_x86_64.whl", hash = "sha256:f5415fb78995644253370985342cd03572ef8620b934da27d77377a2285955bf", size = 20949588, upload-time = "2025-09-09T15:56:59.087Z" },
{ url = "https://files.pythonhosted.org/packages/a6/e4/07970e3bed0b1384d22af1e9912527ecbeb47d3b26e9b6a3bced068b3bea/numpy-2.3.3-cp313-cp313-macosx_11_0_arm64.whl", hash = "sha256:d00de139a3324e26ed5b95870ce63be7ec7352171bc69a4cf1f157a48e3eb6b7", size = 14177802, upload-time = "2025-09-09T15:57:01.73Z" },
{ url = "https://files.pythonhosted.org/packages/35/c7/477a83887f9de61f1203bad89cf208b7c19cc9fef0cebef65d5a1a0619f2/numpy-2.3.3-cp313-cp313-macosx_14_0_arm64.whl", hash = "sha256:9dc13c6a5829610cc07422bc74d3ac083bd8323f14e2827d992f9e52e22cd6a6", size = 5106537, upload-time = "2025-09-09T15:57:03.765Z" },
{ url = "https://files.pythonhosted.org/packages/52/47/93b953bd5866a6f6986344d045a207d3f1cfbad99db29f534ea9cee5108c/numpy-2.3.3-cp313-cp313-macosx_14_0_x86_64.whl", hash = "sha256:d79715d95f1894771eb4e60fb23f065663b2298f7d22945d66877aadf33d00c7", size = 6640743, upload-time = "2025-09-09T15:57:07.921Z" },
{ url = "https://files.pythonhosted.org/packages/23/83/377f84aaeb800b64c0ef4de58b08769e782edcefa4fea712910b6f0afd3c/numpy-2.3.3-cp313-cp313-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:952cfd0748514ea7c3afc729a0fc639e61655ce4c55ab9acfab14bda4f402b4c", size = 14278881, upload-time = "2025-09-09T15:57:11.349Z" },
{ url = "https://files.pythonhosted.org/packages/9a/a5/bf3db6e66c4b160d6ea10b534c381a1955dfab34cb1017ea93aa33c70ed3/numpy-2.3.3-cp313-cp313-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:5b83648633d46f77039c29078751f80da65aa64d5622a3cd62aaef9d835b6c93", size = 16636301, upload-time = "2025-09-09T15:57:14.245Z" },
{ url = "https://files.pythonhosted.org/packages/a2/59/1287924242eb4fa3f9b3a2c30400f2e17eb2707020d1c5e3086fe7330717/numpy-2.3.3-cp313-cp313-musllinux_1_2_aarch64.whl", hash = "sha256:b001bae8cea1c7dfdb2ae2b017ed0a6f2102d7a70059df1e338e307a4c78a8ae", size = 16053645, upload-time = "2025-09-09T15:57:16.534Z" },
{ url = "https://files.pythonhosted.org/packages/e6/93/b3d47ed882027c35e94ac2320c37e452a549f582a5e801f2d34b56973c97/numpy-2.3.3-cp313-cp313-musllinux_1_2_x86_64.whl", hash = "sha256:8e9aced64054739037d42fb84c54dd38b81ee238816c948c8f3ed134665dcd86", size = 18578179, upload-time = "2025-09-09T15:57:18.883Z" },
{ url = "https://files.pythonhosted.org/packages/20/d9/487a2bccbf7cc9d4bfc5f0f197761a5ef27ba870f1e3bbb9afc4bbe3fcc2/numpy-2.3.3-cp313-cp313-win32.whl", hash = "sha256:9591e1221db3f37751e6442850429b3aabf7026d3b05542d102944ca7f00c8a8", size = 6312250, upload-time = "2025-09-09T15:57:21.296Z" },
{ url = "https://files.pythonhosted.org/packages/1b/b5/263ebbbbcede85028f30047eab3d58028d7ebe389d6493fc95ae66c636ab/numpy-2.3.3-cp313-cp313-win_amd64.whl", hash = "sha256:f0dadeb302887f07431910f67a14d57209ed91130be0adea2f9793f1a4f817cf", size = 12783269, upload-time = "2025-09-09T15:57:23.034Z" },
{ url = "https://files.pythonhosted.org/packages/fa/75/67b8ca554bbeaaeb3fac2e8bce46967a5a06544c9108ec0cf5cece559b6c/numpy-2.3.3-cp313-cp313-win_arm64.whl", hash = "sha256:3c7cf302ac6e0b76a64c4aecf1a09e51abd9b01fc7feee80f6c43e3ab1b1dbc5", size = 10195314, upload-time = "2025-09-09T15:57:25.045Z" },
{ url = "https://files.pythonhosted.org/packages/11/d0/0d1ddec56b162042ddfafeeb293bac672de9b0cfd688383590090963720a/numpy-2.3.3-cp313-cp313t-macosx_10_13_x86_64.whl", hash = "sha256:eda59e44957d272846bb407aad19f89dc6f58fecf3504bd144f4c5cf81a7eacc", size = 21048025, upload-time = "2025-09-09T15:57:27.257Z" },
{ url = "https://files.pythonhosted.org/packages/36/9e/1996ca6b6d00415b6acbdd3c42f7f03ea256e2c3f158f80bd7436a8a19f3/numpy-2.3.3-cp313-cp313t-macosx_11_0_arm64.whl", hash = "sha256:823d04112bc85ef5c4fda73ba24e6096c8f869931405a80aa8b0e604510a26bc", size = 14301053, upload-time = "2025-09-09T15:57:30.077Z" },
{ url = "https://files.pythonhosted.org/packages/05/24/43da09aa764c68694b76e84b3d3f0c44cb7c18cdc1ba80e48b0ac1d2cd39/numpy-2.3.3-cp313-cp313t-macosx_14_0_arm64.whl", hash = "sha256:40051003e03db4041aa325da2a0971ba41cf65714e65d296397cc0e32de6018b", size = 5229444, upload-time = "2025-09-09T15:57:32.733Z" },
{ url = "https://files.pythonhosted.org/packages/bc/14/50ffb0f22f7218ef8af28dd089f79f68289a7a05a208db9a2c5dcbe123c1/numpy-2.3.3-cp313-cp313t-macosx_14_0_x86_64.whl", hash = "sha256:6ee9086235dd6ab7ae75aba5662f582a81ced49f0f1c6de4260a78d8f2d91a19", size = 6738039, upload-time = "2025-09-09T15:57:34.328Z" },
{ url = "https://files.pythonhosted.org/packages/55/52/af46ac0795e09657d45a7f4db961917314377edecf66db0e39fa7ab5c3d3/numpy-2.3.3-cp313-cp313t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:94fcaa68757c3e2e668ddadeaa86ab05499a70725811e582b6a9858dd472fb30", size = 14352314, upload-time = "2025-09-09T15:57:36.255Z" },
{ url = "https://files.pythonhosted.org/packages/a7/b1/dc226b4c90eb9f07a3fff95c2f0db3268e2e54e5cce97c4ac91518aee71b/numpy-2.3.3-cp313-cp313t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:da1a74b90e7483d6ce5244053399a614b1d6b7bc30a60d2f570e5071f8959d3e", size = 16701722, upload-time = "2025-09-09T15:57:38.622Z" },
{ url = "https://files.pythonhosted.org/packages/9d/9d/9d8d358f2eb5eced14dba99f110d83b5cd9a4460895230f3b396ad19a323/numpy-2.3.3-cp313-cp313t-musllinux_1_2_aarch64.whl", hash = "sha256:2990adf06d1ecee3b3dcbb4977dfab6e9f09807598d647f04d385d29e7a3c3d3", size = 16132755, upload-time = "2025-09-09T15:57:41.16Z" },
{ url = "https://files.pythonhosted.org/packages/b6/27/b3922660c45513f9377b3fb42240bec63f203c71416093476ec9aa0719dc/numpy-2.3.3-cp313-cp313t-musllinux_1_2_x86_64.whl", hash = "sha256:ed635ff692483b8e3f0fcaa8e7eb8a75ee71aa6d975388224f70821421800cea", size = 18651560, upload-time = "2025-09-09T15:57:43.459Z" },
{ url = "https://files.pythonhosted.org/packages/5b/8e/3ab61a730bdbbc201bb245a71102aa609f0008b9ed15255500a99cd7f780/numpy-2.3.3-cp313-cp313t-win32.whl", hash = "sha256:a333b4ed33d8dc2b373cc955ca57babc00cd6f9009991d9edc5ddbc1bac36bcd", size = 6442776, upload-time = "2025-09-09T15:57:45.793Z" },
{ url = "https://files.pythonhosted.org/packages/1c/3a/e22b766b11f6030dc2decdeff5c2fb1610768055603f9f3be88b6d192fb2/numpy-2.3.3-cp313-cp313t-win_amd64.whl", hash = "sha256:4384a169c4d8f97195980815d6fcad04933a7e1ab3b530921c3fef7a1c63426d", size = 12927281, upload-time = "2025-09-09T15:57:47.492Z" },
{ url = "https://files.pythonhosted.org/packages/7b/42/c2e2bc48c5e9b2a83423f99733950fbefd86f165b468a3d85d52b30bf782/numpy-2.3.3-cp313-cp313t-win_arm64.whl", hash = "sha256:75370986cc0bc66f4ce5110ad35aae6d182cc4ce6433c40ad151f53690130bf1", size = 10265275, upload-time = "2025-09-09T15:57:49.647Z" },
{ url = "https://files.pythonhosted.org/packages/6b/01/342ad585ad82419b99bcf7cebe99e61da6bedb89e213c5fd71acc467faee/numpy-2.3.3-cp314-cp314-macosx_10_13_x86_64.whl", hash = "sha256:cd052f1fa6a78dee696b58a914b7229ecfa41f0a6d96dc663c1220a55e137593", size = 20951527, upload-time = "2025-09-09T15:57:52.006Z" },
{ url = "https://files.pythonhosted.org/packages/ef/d8/204e0d73fc1b7a9ee80ab1fe1983dd33a4d64a4e30a05364b0208e9a241a/numpy-2.3.3-cp314-cp314-macosx_11_0_arm64.whl", hash = "sha256:414a97499480067d305fcac9716c29cf4d0d76db6ebf0bf3cbce666677f12652", size = 14186159, upload-time = "2025-09-09T15:57:54.407Z" },
{ url = "https://files.pythonhosted.org/packages/22/af/f11c916d08f3a18fb8ba81ab72b5b74a6e42ead4c2846d270eb19845bf74/numpy-2.3.3-cp314-cp314-macosx_14_0_arm64.whl", hash = "sha256:50a5fe69f135f88a2be9b6ca0481a68a136f6febe1916e4920e12f1a34e708a7", size = 5114624, upload-time = "2025-09-09T15:57:56.5Z" },
{ url = "https://files.pythonhosted.org/packages/fb/11/0ed919c8381ac9d2ffacd63fd1f0c34d27e99cab650f0eb6f110e6ae4858/numpy-2.3.3-cp314-cp314-macosx_14_0_x86_64.whl", hash = "sha256:b912f2ed2b67a129e6a601e9d93d4fa37bef67e54cac442a2f588a54afe5c67a", size = 6642627, upload-time = "2025-09-09T15:57:58.206Z" },
{ url = "https://files.pythonhosted.org/packages/ee/83/deb5f77cb0f7ba6cb52b91ed388b47f8f3c2e9930d4665c600408d9b90b9/numpy-2.3.3-cp314-cp314-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:9e318ee0596d76d4cb3d78535dc005fa60e5ea348cd131a51e99d0bdbe0b54fe", size = 14296926, upload-time = "2025-09-09T15:58:00.035Z" },
{ url = "https://files.pythonhosted.org/packages/77/cc/70e59dcb84f2b005d4f306310ff0a892518cc0c8000a33d0e6faf7ca8d80/numpy-2.3.3-cp314-cp314-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:ce020080e4a52426202bdb6f7691c65bb55e49f261f31a8f506c9f6bc7450421", size = 16638958, upload-time = "2025-09-09T15:58:02.738Z" },
{ url = "https://files.pythonhosted.org/packages/b6/5a/b2ab6c18b4257e099587d5b7f903317bd7115333ad8d4ec4874278eafa61/numpy-2.3.3-cp314-cp314-musllinux_1_2_aarch64.whl", hash = "sha256:e6687dc183aa55dae4a705b35f9c0f8cb178bcaa2f029b241ac5356221d5c021", size = 16071920, upload-time = "2025-09-09T15:58:05.029Z" },
{ url = "https://files.pythonhosted.org/packages/b8/f1/8b3fdc44324a259298520dd82147ff648979bed085feeacc1250ef1656c0/numpy-2.3.3-cp314-cp314-musllinux_1_2_x86_64.whl", hash = "sha256:d8f3b1080782469fdc1718c4ed1d22549b5fb12af0d57d35e992158a772a37cf", size = 18577076, upload-time = "2025-09-09T15:58:07.745Z" },
{ url = "https://files.pythonhosted.org/packages/f0/a1/b87a284fb15a42e9274e7fcea0dad259d12ddbf07c1595b26883151ca3b4/numpy-2.3.3-cp314-cp314-win32.whl", hash = "sha256:cb248499b0bc3be66ebd6578b83e5acacf1d6cb2a77f2248ce0e40fbec5a76d0", size = 6366952, upload-time = "2025-09-09T15:58:10.096Z" },
{ url = "https://files.pythonhosted.org/packages/70/5f/1816f4d08f3b8f66576d8433a66f8fa35a5acfb3bbd0bf6c31183b003f3d/numpy-2.3.3-cp314-cp314-win_amd64.whl", hash = "sha256:691808c2b26b0f002a032c73255d0bd89751425f379f7bcd22d140db593a96e8", size = 12919322, upload-time = "2025-09-09T15:58:12.138Z" },
{ url = "https://files.pythonhosted.org/packages/8c/de/072420342e46a8ea41c324a555fa90fcc11637583fb8df722936aed1736d/numpy-2.3.3-cp314-cp314-win_arm64.whl", hash = "sha256:9ad12e976ca7b10f1774b03615a2a4bab8addce37ecc77394d8e986927dc0dfe", size = 10478630, upload-time = "2025-09-09T15:58:14.64Z" },
{ url = "https://files.pythonhosted.org/packages/d5/df/ee2f1c0a9de7347f14da5dd3cd3c3b034d1b8607ccb6883d7dd5c035d631/numpy-2.3.3-cp314-cp314t-macosx_10_13_x86_64.whl", hash = "sha256:9cc48e09feb11e1db00b320e9d30a4151f7369afb96bd0e48d942d09da3a0d00", size = 21047987, upload-time = "2025-09-09T15:58:16.889Z" },
{ url = "https://files.pythonhosted.org/packages/d6/92/9453bdc5a4e9e69cf4358463f25e8260e2ffc126d52e10038b9077815989/numpy-2.3.3-cp314-cp314t-macosx_11_0_arm64.whl", hash = "sha256:901bf6123879b7f251d3631967fd574690734236075082078e0571977c6a8e6a", size = 14301076, upload-time = "2025-09-09T15:58:20.343Z" },
{ url = "https://files.pythonhosted.org/packages/13/77/1447b9eb500f028bb44253105bd67534af60499588a5149a94f18f2ca917/numpy-2.3.3-cp314-cp314t-macosx_14_0_arm64.whl", hash = "sha256:7f025652034199c301049296b59fa7d52c7e625017cae4c75d8662e377bf487d", size = 5229491, upload-time = "2025-09-09T15:58:22.481Z" },
{ url = "https://files.pythonhosted.org/packages/3d/f9/d72221b6ca205f9736cb4b2ce3b002f6e45cd67cd6a6d1c8af11a2f0b649/numpy-2.3.3-cp314-cp314t-macosx_14_0_x86_64.whl", hash = "sha256:533ca5f6d325c80b6007d4d7fb1984c303553534191024ec6a524a4c92a5935a", size = 6737913, upload-time = "2025-09-09T15:58:24.569Z" },
{ url = "https://files.pythonhosted.org/packages/3c/5f/d12834711962ad9c46af72f79bb31e73e416ee49d17f4c797f72c96b6ca5/numpy-2.3.3-cp314-cp314t-manylinux_2_27_aarch64.manylinux_2_28_aarch64.whl", hash = "sha256:0edd58682a399824633b66885d699d7de982800053acf20be1eaa46d92009c54", size = 14352811, upload-time = "2025-09-09T15:58:26.416Z" },
{ url = "https://files.pythonhosted.org/packages/a1/0d/fdbec6629d97fd1bebed56cd742884e4eead593611bbe1abc3eb40d304b2/numpy-2.3.3-cp314-cp314t-manylinux_2_27_x86_64.manylinux_2_28_x86_64.whl", hash = "sha256:367ad5d8fbec5d9296d18478804a530f1191e24ab4d75ab408346ae88045d25e", size = 16702689, upload-time = "2025-09-09T15:58:28.831Z" },
{ url = "https://files.pythonhosted.org/packages/9b/09/0a35196dc5575adde1eb97ddfbc3e1687a814f905377621d18ca9bc2b7dd/numpy-2.3.3-cp314-cp314t-musllinux_1_2_aarch64.whl", hash = "sha256:8f6ac61a217437946a1fa48d24c47c91a0c4f725237871117dea264982128097", size = 16133855, upload-time = "2025-09-09T15:58:31.349Z" },
{ url = "https://files.pythonhosted.org/packages/7a/ca/c9de3ea397d576f1b6753eaa906d4cdef1bf97589a6d9825a349b4729cc2/numpy-2.3.3-cp314-cp314t-musllinux_1_2_x86_64.whl", hash = "sha256:179a42101b845a816d464b6fe9a845dfaf308fdfc7925387195570789bb2c970", size = 18652520, upload-time = "2025-09-09T15:58:33.762Z" },
{ url = "https://files.pythonhosted.org/packages/fd/c2/e5ed830e08cd0196351db55db82f65bc0ab05da6ef2b72a836dcf1936d2f/numpy-2.3.3-cp314-cp314t-win32.whl", hash = "sha256:1250c5d3d2562ec4174bce2e3a1523041595f9b651065e4a4473f5f48a6bc8a5", size = 6515371, upload-time = "2025-09-09T15:58:36.04Z" },
{ url = "https://files.pythonhosted.org/packages/47/c7/b0f6b5b67f6788a0725f744496badbb604d226bf233ba716683ebb47b570/numpy-2.3.3-cp314-cp314t-win_amd64.whl", hash = "sha256:b37a0b2e5935409daebe82c1e42274d30d9dd355852529eab91dab8dcca7419f", size = 13112576, upload-time = "2025-09-09T15:58:37.927Z" },
{ url = "https://files.pythonhosted.org/packages/06/b9/33bba5ff6fb679aa0b1f8a07e853f002a6b04b9394db3069a1270a7784ca/numpy-2.3.3-cp314-cp314t-win_arm64.whl", hash = "sha256:78c9f6560dc7e6b3990e32df7ea1a50bbd0e2a111e05209963f5ddcab7073b0b", size = 10545953, upload-time = "2025-09-09T15:58:40.576Z" },
{ url = "https://files.pythonhosted.org/packages/95/12/8f2020a8e8b8383ac0177dc9570aad031a3beb12e38847f7129bacd96228/numpy-1.26.4-cp312-cp312-macosx_10_9_x86_64.whl", hash = "sha256:b3ce300f3644fb06443ee2222c2201dd3a89ea6040541412b8fa189341847218", size = 20335901, upload-time = "2024-02-05T23:55:32.801Z" },
{ url = "https://files.pythonhosted.org/packages/75/5b/ca6c8bd14007e5ca171c7c03102d17b4f4e0ceb53957e8c44343a9546dcc/numpy-1.26.4-cp312-cp312-macosx_11_0_arm64.whl", hash = "sha256:03a8c78d01d9781b28a6989f6fa1bb2c4f2d51201cf99d3dd875df6fbd96b23b", size = 13685868, upload-time = "2024-02-05T23:55:56.28Z" },
{ url = "https://files.pythonhosted.org/packages/79/f8/97f10e6755e2a7d027ca783f63044d5b1bc1ae7acb12afe6a9b4286eac17/numpy-1.26.4-cp312-cp312-manylinux_2_17_aarch64.manylinux2014_aarch64.whl", hash = "sha256:9fad7dcb1aac3c7f0584a5a8133e3a43eeb2fe127f47e3632d43d677c66c102b", size = 13925109, upload-time = "2024-02-05T23:56:20.368Z" },
{ url = "https://files.pythonhosted.org/packages/0f/50/de23fde84e45f5c4fda2488c759b69990fd4512387a8632860f3ac9cd225/numpy-1.26.4-cp312-cp312-manylinux_2_17_x86_64.manylinux2014_x86_64.whl", hash = "sha256:675d61ffbfa78604709862923189bad94014bef562cc35cf61d3a07bba02a7ed", size = 17950613, upload-time = "2024-02-05T23:56:56.054Z" },
{ url = "https://files.pythonhosted.org/packages/4c/0c/9c603826b6465e82591e05ca230dfc13376da512b25ccd0894709b054ed0/numpy-1.26.4-cp312-cp312-musllinux_1_1_aarch64.whl", hash = "sha256:ab47dbe5cc8210f55aa58e4805fe224dac469cde56b9f731a4c098b91917159a", size = 13572172, upload-time = "2024-02-05T23:57:21.56Z" },
{ url = "https://files.pythonhosted.org/packages/76/8c/2ba3902e1a0fc1c74962ea9bb33a534bb05984ad7ff9515bf8d07527cadd/numpy-1.26.4-cp312-cp312-musllinux_1_1_x86_64.whl", hash = "sha256:1dda2e7b4ec9dd512f84935c5f126c8bd8b9f2fc001e9f54af255e8c5f16b0e0", size = 17786643, upload-time = "2024-02-05T23:57:56.585Z" },
{ url = "https://files.pythonhosted.org/packages/28/4a/46d9e65106879492374999e76eb85f87b15328e06bd1550668f79f7b18c6/numpy-1.26.4-cp312-cp312-win32.whl", hash = "sha256:50193e430acfc1346175fcbdaa28ffec49947a06918b7b92130744e81e640110", size = 5677803, upload-time = "2024-02-05T23:58:08.963Z" },
{ url = "https://files.pythonhosted.org/packages/16/2e/86f24451c2d530c88daf997cb8d6ac622c1d40d19f5a031ed68a4b73a374/numpy-1.26.4-cp312-cp312-win_amd64.whl", hash = "sha256:08beddf13648eb95f8d867350f6a018a4be2e5ad54c8d8caed89ebca558b2818", size = 15517754, upload-time = "2024-02-05T23:58:36.364Z" },
]
[[package]]